Abstract

Introduction:

Neurological post-acute sequelae of COVID-19 (neuroPASC) are associated with persistent cognitive dysfunction and quality-of-life decline. We aimed to identify clinical, behavioral and sociodemographic factors associated with neuroPASC symptom burden two years after COVID-19 among individuals without prior neurological disease.

Methods:

In this prospective, observational study, individuals with neuroPASC (n = 102) and controls without symptomatic COVID-19 (n = 74), all without prior neurological, psychiatric, or post-viral conditions, were enrolled between February 2022 and June 2024 across five academic sites. An unsupervised algorithm identified clusters with differing self-reported neurological symptom burden within the neuroPASC group. Functional differences between clusters were evaluated using quality-of-life, neurological and cognitive evaluations. Demographics, behavioral history, comorbidities, and blood biomarkers were compared across clusters and controls. Multivariable logistic regression assessed predictors of neuroPASC severity, including demographics, body-mass-index, Charlson Comorbidity Index, Framingham Risk Score, pre-existing endocrine/metabolic and/or gastrointestinal/hepatobiliary conditions, COVID-19 vaccination prior to infection, hospitalization during acute infection, and cumulative alcohol use.

Results:

Two clusters emerged based on neurological symptom burden, labeled “high-burden” and “low-burden” neuroPASC, reflecting differences in the number and frequency of symptoms. Both clusters had deficits in quality-of-life and cognitive function compared to controls, with greater impairment in high-burden than low-burden neuroPASC. The clusters did not differ by sex, education, tobacco and cannabis use, blood pressure, body-mass-index, HbA1C, days since infection, hospitalization during COVID-19, pre-COVID vaccination rate, antibody-positivity, inflammation, and neurodegeneration biomarkers. The high-burden cluster was older and exhibited higher comorbidity burden and greater cumulative alcohol use compared with the low-burden cluster and controls. Pre-existing endocrine/metabolic and gastrointestinal/hepatobiliary conditions were more common in high-burden (63%) than in low-burden neuroPASC (35%). After adjusting for clinical and demographic factors, these pre-existing conditions remained the only independent predictor of severity, conferring a 3.5-fold increase in the odds of high-burden versus low-burden neuroPASC.

Discussion:

Older age, higher comorbidity burden, greater cumulative alcohol use, and endocrine/metabolic and gastrointestinal conditions, rather than acute COVID-19 severity, were observed in the high-burden neuroPASC cluster. After multivariable adjustment, only pre-existing endocrine/metabolic and/or gastrointestinal/hepatobiliary conditions remained independently associated with high-burden neuroPASC, conferring a 3.5-fold increase in odds and highlighting the need for targeted post-infection monitoring in at-risk patients.

Introduction

Long COVID is a static, relapsing and remitting, or progressive disease that occurs after SARS-CoV-2 infection, with a minimum duration of 3 months (National Academies of Sciences, Engineering, and Medicine, 2024). Neurological post-acute sequelae of COVID-19 (neuroPASC) refer specifically to enduring neurological and cognitive symptoms (Mina et al., 2023), and is associated with persistent cognitive dysfunction and quality-of-life decline (Katz et al., 2023; Kim et al., 2023; Szewczyk et al., 2024). The most affected domains reported in neuroPASC are executive function, memory, attention, and processing speed (Taquet et al., 2021; Guo et al., 2022; Ballouz et al., 2023; Cheetham et al., 2023; Panagea et al., 2025), with some individuals continuing to show cognitive dysfunction 2 years after acute COVID-19 (Ballouz et al., 2023). Cognitive recovery usually parallels the resolution of neurological symptoms (Guo et al., 2022; Ballouz et al., 2023; Cheetham et al., 2023).

Several studies have investigated risk factors for long COVID. Overall, older age, certain comorbidities (particularly cardiovascular disease), hospitalization during acute infection, higher body mass index (BMI), and female sex have consistently been reported as risk factors for post-acute sequelae of COVID-19 (Pilotto et al., 2021; Sudre et al., 2021; Tsampasian et al., 2023; Romero-Ibarguengoitia et al., 2024; Alcalde-Herraiz et al., 2025; Hou et al., 2025). In contrast, relatively few studies have specifically examined risk factors associated with persistent neurological symptoms, and findings across these studies have been variable or cohort-specific (Frontera and Simon, 2022; Li et al., 2023; Cahan et al., 2024; Choudhury et al., 2025; Shil et al., 2025). Notably, many studies recruit participants from specialized clinics, potentially excluding individuals with mild or transient symptoms who are less likely to seek specialized care (Exley et al., 2025). Pre-existing major neurological or psychiatric conditions also pose potential confounders, as they can independently affect cognition, quality-of-life and neurological function, and evidence suggests that such symptoms can worsen within two to three years following COVID-19 hospitalization (Taquet et al., 2024) and potentially even after mild-to-moderate infection (Jose, 2024). Moreover, most neurocognitive assessments in long COVID research have focused on previously hospitalized individuals, many of whom required mechanical ventilation as part of their treatment (Panagea et al., 2025). Consequently, the risk factors associated with neuroPASC symptom burden among individuals without prior neurological or psychiatric disorders—and without the confounding influence of mechanical ventilation—remain poorly understood.

The multi-site COVID-Brain Advanced Imaging Network (COVID-BRAIN) (COVID-BRAIN Project, 2021) seeks to elucidate the long-term effects of SARS-CoV-2 infection on the brain through advanced magnetic resonance imaging (MRI), standardized neurological and neuropsychological assessments, and blood-based biomarkers. The neuroPASC group in this study consisted of individuals with no prior history of chronic neurological or active psychiatric disorders, who had not required mechanical ventilation during acute COVID-19, and who continued to experience neurological symptoms two years post-infection. Therefore, the COVID-BRAIN cohort allowed us to examine factors associated with neuroPASC symptom burden without the confounding influence of pre-existing major neurological or psychiatric conditions and from complications associated with mechanical ventilation. Using an unsupervised cluster analysis of self-reported neurological symptoms, we defined neuroPASC sub-groups based on disease burden. We then evaluated daily functioning in these sub-groups using standardized quality-of-life (fatigue, sleep, depression, anxiety) surveys, a structured neurological examination and a centralized cognitive battery comprised of attention, working memory, processing speed and memory assessments to determine if these standardized measures corroborated self-reported disease burden. Finally, we investigated the factors associated with neuroPASC disease burden, including sex, age, body-mass-index (BMI), need for hospitalization and oxygen therapy during acute infection, behavioral history, social determinants of health, pre-existing comorbidities, presence of nucleocapsid antibodies, and blood biomarkers of inflammation, neurodegeneration and amyloid burden.

MethodsParticipants

All procedures were approved by the Institutional Review Board: Human Subjects Committee of the University of Minnesota under a central IRB protocol and informed consent was obtained from all participants. COVID-BRAIN (COVID-BRAIN Project, 2021) participants were enrolled between February 2022 and June 2024 across five sites (University of Minnesota, Mayo Clinic Rochester, Massachusetts General Hospital, Johns Hopkins University, Houston Methodist Research Institute), using the following inclusion criteria:

Age 18 years or older;

Controls: Individuals who had no known SARS-CoV-2 infection;

NeuroPASC group: Individuals who had PCR, antibody or antigen confirmed COVID-19 and presented with neurological symptoms in the 6 months after infection, continued to show at least one post-COVID neurological symptom and fit one of the following criteria during the acute phase of the infection: ambulatory with no or mild symptoms, hospitalized but no oxygen therapy, or hospitalized and on oxygen administered via a nasal cannula, mask or non-invasive ventilation, i.e., individuals with WHO Ordinal Scale scores 0–5 (Rubio-Rivas et al., 2022);

English or Spanish speaking (based on self-stated primary language);

Clear of any contraindications for MRI: including but not limited to claustrophobia, unable to remain still in an MRI scanner for more than 30 min, presence of paramagnetic substances or pacemakers in body, weight over 300 lbs.

Participants were excluded if they had pre-existing chronic neurological conditions (e.g., neurodegenerative disorders, demyelinating and inflammatory central nervous system disorders, cerebrovascular disease and epilepsy), active psychiatric illness (e.g., bipolar disorder, schizophrenia and active substance use disorder excluding cannabis), end-stage renal disease, end-stage liver disease, stroke, brain tumor, brain infection, or traumatic brain injury with loss of consciousness, were diagnosed with another post-viral syndrome or Chronic Fatigue Syndrome before COVID-19, or required mechanical ventilation during hospitalization (WHO Ordinal Scale scores 6–7) (Rubio-Rivas et al., 2022) to avoid the confounding effects of invasive treatment. Participants with well-controlled, treated, or remitted psychiatric conditions—including depression, anxiety, and attention deficit hyperactivity disorder—were allowed to participate.

Screening for the study involved two steps (Figure 1): interested participants underwent an online screening (Step 1) via REDCap through the study webpage that described study goals and basic eligibility criteria.1 Participants who qualified through the online survey were automatically connected to the coordinator at the nearest enrollment site and underwent a detailed phone screening (Step 2). A total of 117 participants with self-reported neuroPASC (35 of them hospitalized during COVID-19) and 81 control participants were enrolled. Nineteen participants were withdrawn after enrollment (Figure 1), and a total of 179 individuals completed a comprehensive study visit that included MRI acquisition. Three participants were subsequently excluded after the study visit upon discovery that they met exclusion criteria or presented incidental MRI findings likely unrelated to COVID-19. In total, we assessed 176 individuals (neuroPASC n = 102, controls n = 74; Table 1). All individuals with neuroPASC reported at least one neurological symptom persisting for three or more months, consistent with the long COVID definition developed by the National Academies of Sciences, Engineering, and Medicine (2024).

CONSORT diagram. Flow-chart showing how the COVID-BRAIN cohort of individuals with neuroPASC and controls was built. Most interested participants underwent an online screening via REDCap through the study webpage that described study goals and basic eligibility criteria (covidbrainstudy.umn.edu). Interested participants who qualified through the online survey were automatically connected to the coordinator at the nearest enrollment site for a phone screening.

VariableControl participants (n = 74)Participants with NeuroPASCp value neuroPASC vs controlp value three group comparison (Control, low-burden, high-burden neuroPASC)Total group (n = 102)Low-burden neuroPASC (n = 75)High-burden neuroPASC (n = 27)Site NUMN3237289Mayo2429245MGH1722139JHU11174Houston0330Age, median (IQR), y43 (28–59)50 (35–59)45 (32–58)56 (47–64)0.1a0.007b, 0.3c, 0.003d, 0.005eFemale, No. (%)47 (63%)75 (74%)56 (75%)20 (74%)0.1f0.5g, 0.5h, >0.99iEducation, median (IQR), y16 (16–18)16 (14–18)16 (14–18)16 (14–18)0.2a0.4b, 0.3c, 0.3d, 0.3eDays since infection, median (IQR), dNA781 (504–1,005)763 (492–950)859 (583–1,198)NA0.1aHospitalized during COVID-19, No. (%)NA25 (25%)16 (21%)9 (33%)NA0.3iNeeded oxygen therapy, No. (%)NA22 (22%)14 (19%)8 (30%)NA0.3iVaccinated for COVID-19, No. (%)71 (96%)94 (92%)68 (91%)26 (96%)0.5f0.7g, 0.7h, 0.7iVaccinated for COVID-19 before infection, No. (%)NA40 (39%)30 (40%)10 (37%)NA0.8iBMI, median (IQR)25 (23–32)28 (25–36), n = 10129 (25–36)28 (26–33), n = 260.006j0.01k, 0.01l, 0.3m, 0.5nHbA1C, %, median (IQR)5.2 (5–6), n = 665.3 (5–6), n = 795.2 (5–6), n = 615.3 (5–6), n = 180.1j0.1k, 0.1l, >0.99m, 0.3nBlood pressure (mmHg), median (IQR)Systolic121 (112–130)122 (110–132)122 (110–131)123 (111–134)0.6j0.6k, 0.8l, 0.6m, 0.6nDiastolic78 (72–84)78 (71–83)78 (71–83)77 (75–82)0.8j0.9k, 0.8l, 0.8m, 0.8nCCI, median (IQR)0 (0–2)1 (0–2)0 (0–1)2 (0–3)0.3a0.03b, 0.5c, 0.01d, 0.02eFRS, median (IQR)3 (1–7), n = 735 (1–10), n = 1014 (1–10), n = 747 (3–11)0.1a0.02b, 0.3c, 0.01d, 0.02ePositive for SARS-CoV-2 nucleocapsid No. (%)37 (51%), n = 7294 (93%), n = 10170 (95%), n = 7424 (89%)<0.001f<0.001g, 0.001h, 0.4i

Demographics and clinical characteristics of participants with neuroPASC and controls.

Participants with neuroPASC were empirically grouped in two distinct clusters, low-burden and high-burden neuroPASC (Figure 2), based on the self-reported burden of neurological symptoms that were ongoing at the time of visit. Group comparisons were conducted using the Benjamini–Hochberg correction for multiple testing (false discovery rate, FDR < 0.05) across the following contrasts: controls vs. low-burden neuroPASC, controls vs. high-burden neuroPASC, and low-burden vs. high-burden neuroPASC. Sample sizes are indicated in cases with missing data. CCI, Charlson Comorbidity Index; FRS, 10-Year Framingham Risk score.

b

Kruskal-Wallis rank sum test.

c

Dunn adjusted pairwise comparison, control vs low-burden neuroPASC.

d

Dunn adjusted pairwise comparison, control vs high-burden neuroPASC.

e

Dunn adjusted pairwise comparison, low-burden vs high-burden neuroPASC.

g

Fisher’s exact test, control vs low-burden neuroPASC.

h

Fisher’s exact test, control vs high-burden neuroPASC.

i

Fisher’s exact test, low-burden vs high-burden neuroPASC.

j

ANCOVA 1 degree of freedom F-test, adjusted for age and sex.

k

ANCOVA 2 degree of freedom F-test, adjusted for age and sex.

l

Pairwise comparison, control vs low-burden neuroPASC, adjusted for age and sex.

m

Pairwise comparison, control vs high-burden neuroPASC, adjusted for age and sex.

n

Pairwise comparison, low-burden vs high-burden neuroPASC, adjusted for age and sex.Bold values were used to highlight significant p-values (p < 0.05).

Self-reported neuroPASC symptoms

NeuroPASC symptoms obtained by a structured interview included altered mental status (reduced level of awareness, confusion, brain fog), headaches, behavioral change, speech disturbances, muscle weakness, myalgia, paresthesia/limb pain, dysphagia, loss of smell, loss of taste, photophobia, visual disturbances, and seizures. The list was adapted from the neurological symptom checklist in the COVID-Neuro Network case record form (Brain Infections Global, 2020), which was developed through expert consensus to facilitate standardized data collection from patients with neurological complications of COVID-19. Participants reported when they experienced the symptom, whether the symptom was ongoing (experienced within the past 14 days), and weekly symptom frequency at its most severe. NeuroPASC symptoms were defined as those neurological symptoms that were either: (1) experienced during and persisting beyond the acute phase; or (2) newly emergent after the acute phase. The case report form that formed the basis of the REDCap symptom survey, which was filled by neurologist co-investigators at each site, is provided in Supplementary Forms (“COVID-BRAIN NEUROLOGICAL SYMPTOMS”).

Clinical assessment and neurological examination

Quality-of-life questionnaires included Modified Fatigue Impact Scale (MFIS), Pittsburgh Sleep Quality Index (PSQI), Patient Health Questionnaire (PHQ-8), and General Anxiety Disorder (GAD-7). Self-reported medical history was recorded, including endocrine/metabolic, gastrointestinal/hepatobiliary, psychiatric, neurological, cardiovascular, respiratory, musculoskeletal, ocular/vision, renal/urinary, dermatological, blood, reproductive, tumor/cancer, systemic or other conditions. A structured neurological exam (see Supplementary Forms “COVID-BRAIN NEUROLOGICAL EXAM”) was administered, including meningeal signs, mental status, cranial nerves, motor functions, sensation, coordination, reflexes and gait (Campbell, 2012).

Neurocognitive assessment

A battery of neuropsychological tests was administered by the same examiner (KB), to maintain evaluator consistency, and via a HIPAA compliant video platform to minimize in-person contact time. Virtual neuropsychological evaluations have become increasingly common and have been found to provide a valid and reliable assessment of cognitive function in multiple patient populations (Iiboshi et al., 2020; Watt et al., 2021; Gallagher et al., 2023). Global cognition, attention, working memory, processing speed, executive functions, memory, language, and visual spatial skills were assessed using: Montreal Cognitive Assessment (MoCA) Blind; WAIS-IV Digit Span subtest; Stroop Color-Word Interference Test; Symbol Digit Modalities Test – oral version (SDMT); Controlled Oral Word Association Test (COWAT): FAS; Animal Fluency; Delis- Kaplan Executive Function Systems (DKEFS) Verbal Fluency: Category Switching subtest (fruit/furniture); Hopkins Verbal Learning Test-Revised (HVLT-R); Brief Visuospatial Memory Test-Revised (BVMT-R); Neuropsychological Assessment Battery (NAB) Naming Test; and Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) Line Orientation subtest. The virtual assessment precluded inclusion of the Trail Making Test that shows impairments after COVID-19 (Miskowiak et al., 2021; Yesilkaya et al., 2021); therefore the DKEFS Category Switching subtest was used as an alternative test of cognitive flexibility. Validated Spanish versions were used for one participant whose primary language was Spanish. The following variables were analyzed based on literature documenting deficits in attention, working memory, processing speed, and memory in neuroPASC (Cecchetti et al., 2022; Kay et al., 2022; Vannorsdall et al., 2022; Díez-Cirarda et al., 2023): MoCA, SDMT, Stroop Color-Word, Stroop Interference, FAS, DKEFS Category Switching (Fruit/Furniture), Digit Span backward, Digit Span combined, HVLT-R total recall, HVLT-R delayed recall (DR), BVMT-R total recall, BVMT-R DR.

Blood biomarker analyses

Details on how plasma SARS-CoV-2 nucleocapsid antibodies, hemoglobin A1c (HbA1C), high-sensitivity C-reactive protein (hsCRP), interleukin (IL)-1, IL-6, tumor necrosis factor-alpha (TNF-α), Aβ42/40 ratio, plasma phosphorylated tau-181 (pTau181), neurofilament light (NfL), and glial fibrillary acidic protein (GFAP) were quantified and how presence of APOE ε4 allele was determined are in Supplementary Methods. For data consistency, blood specimens collected at each site were shipped to the Advanced Research and Diagnostic Laboratory (ARDL, https://med.umn.edu/pathology/research/ardl) at the University of Minnesota and centrally analyzed using the methods detailed in Supplementary Methods. ARDL established a specimen collection protocol, prepared training materials for site personnel and shipped sample collection kits and bulk supplies to sites.

Cluster analysis of neuroPASC symptoms

To identify clusters of participants based on self-reported symptom burden, the highest weekly frequency was used, with an ordinal scale: not present = 0, less than once per week = 1, 1–2 times per week = 3, 3–4 times per week = 4, 5–6 times per week = 5, every day = 6. The gap between 1 and 3 was intentionally widened to reflect the substantive difference between sporadic and weekly symptom occurrence (“less than once per week” vs. “1–2 times per week”). This non-linear encoding preserves ordinal rank while emphasizing clinically meaningful transitions for clustering analysis. As a sensitivity analysis, we repeated the clustering analysis using a linear encoding of symptom frequency (0–5) to confirm that the identified clusters were not driven by the deliberate introduction of a widened frequency gap. A heatmap of subject-level symptom frequency is provided in Supplementary Figure 1. Because seizures following COVID-19 infection were reported by only one participant, this symptom was excluded from the cluster analysis. We performed unsupervised K-means clustering on symptom frequency data after standardizing each symptom to z-scores across participants. Clustering was repeated using 1,000 random starts and 10 iterations per run using the “kmeansruns” function in the “fpc” package in R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria). We used the “NbClust” R package (Charrad et al., 2014), evaluating 30 cluster validity indices including Silhouette, Calinski–Harabasz, Davies–Bouldin, Dunn, and others. The optimal number of clusters was determined by majority vote across all indices and across 2 to 10 clusters.

Statistical analyses

Statistical analyses were performed in R. To evaluate group differences in neurocognitive assessment, quality-of-life scales, behavior history, BMI, blood pressure and blood biomarkers we used ANCOVA followed by pairwise tests for group comparisons, while adjusting for age and sex differences. Given the known association between BMI and hsCRP (Choi et al., 2013), we additionally included a group × BMI interaction term in the model when assessing group differences in hsCRP. We applied logarithmic transformation with a base of 10 to hsCRP measures and inverse transformations to IL-1 and IL-6 measures, when assessing group differences, due to high skewness in these measures. For clinical relevance, we present non-transformed values in tables. Cohen’s d effect sizes were computed for the pairwise comparisons of neurocognitive and quality-of-life measures using covariance-corrected residuals, where linear regression was used to adjust for effects of age and sex. For other continuous variables we used Kruskal-Wallis followed by Dunn’s adjusted pairwise comparisons. Group differences in categorical variables were assessed using Fisher’s exact test. We used the Benjamini-Hochberg false discovery rate (FDR, p < 0.05) to account for multiple testing. Relevant statistical tests are described as footnotes in Tables 1, 2 and Supplementary Tables.

VariableControl participants (n = 74)Participants with NeuroPASCp value neuroPASC vs controlp value three group comparison (Control, low-burden, high-burden neuroPASC)Total group (n = 102)Low-burden neuroPASC (n = 75)High-burden neuroPASC (n = 27)Behavior historyTobaccoCurrent use, No. (%)3 (4%)5 (5%)2 (3%)3 (11%)>0.99a0.7b, 0.5c, 0.3dPack-years, median (IQR)0 (0–0)0 (0–0.1)0 (0–0)0 (0–2.3)0.2e0.1f, 0.7g, 0.1h, 0.1iAlcoholCurrent use, No. (%)56 (76%)64 (63%)51 (68%)13 (48%)0.1a0.4b, 0.04c, 0.2dPast use, No. (%)64 (86%)89 (87%)66 (88%)23 (85%)>0.99a>0.99b, >0.99c, >0.99dAge started drinking, median (IQR)19 (18–21)18 (16–21)18 (17–21)16 (14–18.5)0.04j<0.001k, 0.2l, <0.001m, 0.001